The Effect of Joint Auditor Pair On Timely Loss Recognition: Evidence From Impairment Tests

The Effect of Joint Auditor Pair on Timely Loss Recognition

Evidence from Impairment Tests

Abstract: As some regulators view joint audit as a way to enhance audit quality, French law already requires two (joint) auditors. We examine the effect of auditor pairs on a key mechanism ensuring timely loss recognition: impairment tests. Impairment tests rely on unverifiable fair value estimates and are often manipulated by managers. In a simple game theory model, assuming that Big 4 auditors face higher reputation and litigation costs than non-Big 4, we demonstrate that pairs of Big 4 auditors are likely to lead to the prisoner’s dilemma solution, according to which conditional conservatism is too low. Conversely, pairs of Big 4 and non-Big 4 auditors increase Big 4’s incentives to force firms to book timely impairments. From a sample of French listed firms (SBF 120) from 2006 to 2009, we provide evidence that: (1) Big 4–Small auditor pairs book more timely impairments; (2) Big 4–Small auditor pairs manage less impairment tests’ transparency; (3) Big 4–Small auditor pairs are more conditionally conservative using the Basu (1997)’s measure of conservatism; (4) Big 4– Small auditor pairs are not associated with lower level of earnings quality as proxied by abnormal accruals.

Keywords: Joint Audit – Conservatism – Timely Loss Recognition – Impairment Test – Prisoner’s Dilemma – Audit Quality

I. INTRODUCTION

Auditors’ inability to prevent the 2008 financial crisis has fueled suspicions that such firms lack “the requisite independence, expertise and incentives to construct the promised ‘true’ and ‘fair’ account of corporate affairs” (Sikka 2009, 868). Solutions aiming to improve the capacity of audit firms to detect and prevent corporate bankruptcies, frauds and failures are under review by different regulatory authorities around the world. In Europe, the European Commission proposed the use of joint audits to improve audit quality and reduce audit market concentration in its Green Paper published in 2010. Similar initiatives have been proposed by regulatory authorities in UK, India and China. Such proposals result in widespread debate over the benefit and cost of joint audit.

Advocates of joint audit claim that it can improve audit quality. DeAngelo (1981) defines audit quality as the market-assessed joint probability that a given auditor will both discover a breach in an accounting system and report the breach. Thus, audit quality is determined by the auditor’s ability to detect material misstatements (auditor competence) and the auditor’s willingness to report discovered material misstatements (auditor independence). Joint audit may affect both components of audit quality. From the competence perspective, having another auditor review the work could lead to a higher chance of discovering the problem than having only one auditor, provided that both auditors take reasonable efforts in the audit. From the independence perspective, auditor independence may improve due to the following reasons: (1) There is lower economic bonding between auditor and client because the audit and consulting fees are shared between two auditors (Mazars 2010; Zerni et al. 2010). (2) Collusion is less likely to appear because it is more difficult to bribe two auditors. As long as the benefits to take corrective actions exceed the costs for one auditor, the problem will be reported and corrected (Zerni et al. 2012, 4). (3) Joint auditors can rotate at different time period, therefore they are expected to increase auditors’ independence (rotation of one firm off the audit), while ensuring continuity by preserving the knowledge of auditors on the auditee (one auditor maintains the relationship) (Carcello and Nagy 2004).

However, opponents of the joint audit regime argue that it is costly to implement and suffers from two main theoretical deficiencies. (1) Empirical evidence in Denmark suggests that the costs of joint audits may on average outweigh its perceived benefits by firms since the switch from mandatory joint audit regime to the voluntary joint audit regime in 2004 led 76% of listed firms to choose only one auditor instead of two (Thinggaard and Kiertzner 2008). (2)

The joint audit regime may create a free-rider problem: one auditor may rely on the other auditor work and may not provide the appropriate level of effort. (3) Organization costs and insufficient cooperation may appear under the joint audit regime in a context of fierce competition to increase market share. Inappropriate cooperation during the audit could lead to insufficient information exchange between the two auditors and in fine lower audit quality (Neveling 2007).

Policy makers have the responsibility to assess the cost-benefit trade-off of the joint audit regime in the perspective of improving financial statement quality. Empirical evidence from Denmark (Thinggaard and Kiertzner 2008) may only reflect the trade-off for auditees’ and not may not consider all potential benefits as Zerni et al. (2012) provide evidence of a positive relation between joint audits and general attributes of audit quality.

In this debate around the joint audit regime and the upcoming European regulation, two questions arise: (1) Should a firm opt for one or two auditors? (2) If a firm chooses two auditors, how many Big 4 auditors, if any, should be selected? We examine the second question. In France, publicly listed companies preparing consolidated financial statements are already required to be audited by at least two unrelated auditors offering a unique opportunity to assess the conditions for effectiveness of the joint audit, which could become an option for other European public-interest entities in the future. This requirement is more complex than the typical choice between Big 4 and non-Big 4 auditors prevailing in most developed countries. Our objective is to examine the relations between auditor pairs, and the quality of audit, in particular the degree of conservatism of audited accounts. Audit quality is the consequence of two effects: independence and competence (knowledge). Theoretical arguments supporting joint audits suggest that it could particularly improve independence of auditors.

In terms of independence, Big 4 auditors may bear a much higher risk when they are paired with non-Big 4 auditors rather than with another Big 4 auditor. In the Big 4 – non-Big 4 pair, the cost of litigation and loss of reputation in case of an audit failure is largely borne by Big 4 auditors whereas such costs are equally shared in the Big 4–Big 4 pair, both having deep pockets. A court decision in France for the company Marionnaud, which was audited by a Big 4 auditor (KPMG) paired with a non-Big 4 auditor (Cofirec), illustrates how litigation risks fall on the Big 4 auditor (AMF 2007):1 KPMG was the only audit firm that was held

1

The AMF (‘Autorité des marchés financiers’ – ‘Financial Market Authority’) is the French equivalent of the SEC.

responsible in 2007 for misstatements that occurred from 2002 to 2004. The AMF acknowledged that the small auditor “didn't have the resources nor the possibility” to detect the irregularities in its cross review. Due to the unproportionately high risk borne by the Big 4 auditors when paired with a small auditor, they have better incentives to enforce appropriate financial reporting, which could result in higher auditor independence. Based on these arguments, the Big 4–Big 4 pair has most likely higher auditor competence but not necessarily implies higher independence. As a result, the overall effect of auditor pairs on audit quality is not clear and remains an empirical question.

We assess the consequences of the different auditor pairs on a central qualitative characteristic of financial statements: conditional conservatism, i.e., incorporating economic losses into earnings when they occur. Conditional conservatism also known as timely loss recognition or news-dependent conservatism consists in writing down book values and decreasing income under sufficiently adverse circumstances. Conversely, book value is not written up when circumstances are favorable. In international accounting standards and U.S. GAAP frameworks, impairment tests are crucial to guarantee timely loss recognition, as impairment tests ensure that assets are not carried at more than their economic value, also known as the recoverable value (IASB 2004). International Financial Reporting Standards require that an impairment loss should be recognize whenever the recoverable amount is below the carrying amount (IAS 36§59). The implementation of impairment tests usually rely on valuation models and involves “significant judgment” from managers (Petersen and Plenborg 2010, 420). Ramanna (2008), Li et al. (2011), Li and Sloan (2011) and Ramanna and Watts (2012) demonstrate that goodwill impairments tend to be manipulated by managers because the procedure relies on unverifiable fair value estimates. Managers are also likely to be more or less transparent regarding the subjective valuation assumptions used in order to conduct impairment tests. Auditors play a central role in maintaining the objectivity and transparency of impairment tests and can take corrective actions to force firms to recognize economic impairments when they occur and to become more transparent.

Our research question is to determine the association between auditor pairs and financial reporting quality using impairment-testing management. We suggest testing the association between auditor pairs and impairment tests for four reasons: (1) objective impairment tests require a truly independent review from the auditors, whereas other measures of audit quality may capture to a larger extent the simultaneous effects of independence and competence; (2) impairment tests play a key role in ensuring timely loss recognition, one of the main property

of financial statement quality; (3) they are usually suspected of manipulation as they rely on fair value estimates most of the time; (4) they are notoriously related to public firms’ largest individual asset for which fair value estimates is required, i.e. goodwill.2

In order to derive testable hypotheses, we rely on a simple game theory model that takes into account strategic interactions among joint auditors. We assume that reputation and litigation costs are higher for Big 4 auditor than for non-Big 4 auditors because Big 4 auditors are international firms that have more to lose and are usually perceived to have deeper pockets than non-Big 4 firms (Gonthier-Besacier and Schatt 2007; Jong-Hag et al. 2008). We show that a Big 4 auditor paired with another Big 4 auditor, having similar reputation and litigation costs associated with the avoidance of economic impairments, leads to the prisoner’s dilemma-solution, according to which “do nothing” is the dominant strategy. In this case, we demonstrate that (Hypothesis 1) Big 4 pairs should lead to a socially suboptimal equilibrium, according to which economic impairments are not recognize (low level of conditional conservatism) and impairment tests are less transparent. On the opposite, we argue that a Big 4 auditor paired with a non-Big 4 auditor bears a large part, if not all, of the reputation and litigations costs associated with the risk of failure of the non-Big 4 auditor. The dominant strategy for the Big 4 auditor would be to take corrective actions to ensure that impairment tests are objective (high level of conditional conservatism) and transparent. Therefore, (Hypothesis 2) Big 4 auditors paired with non-Big 4 auditors should be more likely to lead to the recognition of economic impairments and more transparent impairment tests. We examine empirically these theoretical predictions.

Our sample consists of all non-financial French firms composing the SBF 120 index (120 largest market cap listed in Paris) from 2006 to 2009. We demonstrate that: (1) Big 4–Big 4 auditor pairs are less likely to impair assets when performance is low and are more likely to impair assets independently from performance, whereas Big 4–non-Big 4 auditor pairs are more likely to impair assets when performance is low and less likely to impair assets independently from performance. This result suggests that the characteristic of timely loss recognition is improved with a Big 4–non-Big 4 auditor pair. (2) Using a Transparency Score, impairment tests of firms audited by Big 4–Big 4 auditor pairs become less transparent when firms book an impairment whereas transparency increases for firms audited by Big 4–non-Big 4 auditor pairs. Big 4–Big 4 auditor pairs reduce the level of transparency when firms

2

From 2006 to 2009, goodwill represents on average 27 % of total assets of the 120 French largest listed firms (SBF 120) composing our sample.

recognize impairments. (3) Big 4–Small auditor pairs are associated with higher levels of unconditional conservatism using the market-to-book ratio as a proxy for unconditional conservatism and higher levels of conditional conservatism using the Basu (1997)’s model of conservatism. (4) Big 4–Big 4 auditor pairs are not related with higher earnings quality – defined as lower income-increasing abnormal accruals – than Big 4–non-Big 4 auditor pairs.

Overall, our results are consistent with the predictions of our game theory model. When Big 4 auditors are paired with non-Big 4 auditors, they are more likely to force firms to conduct objective impairment tests because Big 4 auditors bear a large part of reputation and litigation costs.

We contribute to the literature at two levels. (1) We deepen the understanding of the consequences of the joint audit requirement on financial statements’ quality in the perspective of improving audit quality and the upcoming audit reforms around the world. Strategic interactions between joint auditors matter in terms of financial statements’ quality. To the best of our knowledge, this paper is the first to use game theory to study joint audit. (2) We challenge the idea that two Big 4 auditors necessarily improve financial statements’ quality. Considering strategic interactions between joint auditors, we provide evidence that the relation between joint auditors and financial statement’s quality, in particular conservatism, may be more complex than two Big 4 are better than one Big 4 which is better than none.

The remainder of this paper is organized as follows. We provide a description of the audit market in France in section II, and review the related literature in section III. Our game theory model and our hypotheses are presented in section IV. We show our data, empirical analysis and findings in section V before concluding in section VI.

II. THE AUDIT MARKET IN FRANCE

Since 1966 in France, public firms must be audited by (at least) two distinct auditors that share the audit process. This joint audit requirement has been maintained over the years, although threatened by the European regulation introducing consolidated financial reporting in 1984, and was finally reiterated by the “French Security Law” in 2003 that followed the Enron scandal.3 Auditors have a six-year mandate and face (for mandates of listed firms) a

3

French Financial Security Law (2003). « Loi No 2003-706 du 1 août 2003 de sécurité financière, version consolidée au 1er avril 2006 », available at http://www.legifrance.gouv.fr. Francis et al. (2009, 38) also provide details on the specificities of the audit market in France.

compulsory (partner) rotation after each mandate. Joint auditors share the workload and associated fees in conducting the audit process according to quantitative criteria (e.g., number of estimated hours) and qualitative criteria (e.g., expertise required). Therefore, the joint audit is not a double audit where auditors would duplicate its counterpart’s work. Joint auditors must sign a single audit report, i.e., agree on the same report independently from each other, and are jointly liable for the issued audit opinion. The French Security Law also requires that each joint auditor verifies the work undertaken by the other auditor leading to the joint audit report. As a consequence, auditors share the audit work but have a legal obligation to review their counterpart’s work. Joint audit is viewed as an option by the European Commission to restore confidence in the financial statement of companies after the 2008 financial crisis. As Michel Barnier, Internal Market and Services Commissioner, explains “[The European Commission]’s proposals4 address the current weaknesses in the E.U. audit market, by eliminating conflicts of interest, ensuring independence and robust supervision and by facilitating more diversity in what is an overly concentrated market, especially at the top-end.” One of the main arguments put forward by the European commission for the joint audit regime is the willingness to allow new “Big” audit firms to emerge.

France presents G8 lowest concentration level in the audit sector, where Big 4 auditors shared only 61% of the market revenues in 2007 compared to 91% for the other G8 countries.5 The consequence of the joint audit rule is that, even if Big 4 companies dominate the audit market, smaller audit firms also have significant position on the market. Indeed, 53% of our sample that covers the 2006-2009 period and represents the 120 non-financial largest firms by market cap, are audited by at least one non-Big 4 auditor.6 Non-Big 4 auditors can be classified in two sub-groups: (1) Tier-one non-Big 4 auditors having considerable revenues, several listed-firm mandates and belong to an international network, i.e., Mazars, Grand Thornton and BDO; (2) Tier-two non-Big 4 auditors, having much smaller revenues, only one or few mandates of listed firms and are mainly French players (e.g., AEG Finance, Cofirec, Dauge & Associés, Didier Kling & Associés). From 2006 to 2009, small auditors sharing a joint audit mandate with a Big 4 auditor did not complete an important part of the audit process neither critical aspects of the mission most of the time.7 Nonetheless, the legal 4 (European_Commission 2011a, 2011b) 5 http://www.gti.org/Press-room/Press-archive/2007/G8-audit-concentration.asp 6

See Descriptive Statistics in Table 4, Panel B.

7

In February 2012, the oversight board of the legal audit in France (the “Haut Conseil du Commissariat aux Comptes” also called “H3C”) published a report criticizing this unequal share of work between joint auditors. The H3C urged audit firms to share the audit work equally. See http://www.h3c.org/textes/Avis%202012-01.pdf

audit of listed companies for smaller auditors represents a good marketing material, increasing visibility and legitimacy allowing these smaller firms to sell advisory missions to other clients.

III. OVERVIEW OF RELATED LITERATURE

Analytical work about joint audit is scarce. To our best knowledge, the only model designed to evaluate the impact of joint audit is Deng et al. (2012). Their model aims to assess the effect of joint audit on audit fees, audit evidence precision and auditor independence. They compare the situation for Single Big-Firm Auditor (regime B), Two Big-Firm Auditors (regime BB) and One Big-Firm Auditor paired with one Small-Firm Auditor (regime BS). The results suggest that audit evidence precision is the same for regimes B and BB, but lower for regime BS as the small firm free ride on the big firm. In addition, joint audit lowers auditor independence for both regimes BB and BS. Although buying off two auditors is more expensive under joint audit, joint audits provide companies with an opportunity of internally shopping for a favorable audit opinion from its two auditors and thus lead to a higher level of ex post earnings management. In terms of audit fees, the BB regime would result in lower audit fees than the B regime due to the convexity assumption of the resource cost function (one audit firm doing all the work under a completion time constraint may experience a higher cost than if the work was split between two firms). The audit fee for the BS regime would be lower than under the B regime only if the big firm and the small firm had similar technological efficiency or if the big firm would bear a sufficiently large proportion of misstatement cost. In general, the results indicate that, in contrast to common views, joint audit does not necessarily improve auditor competence or independence due to free-riding and internal opinion shopping. Audit fees are also not necessarily higher compared with single audit.

Francis et al. (2009) analyze the consequences of France’s joint audit requirement on earnings quality on a sample of 261 firm-year observations and find that Big 4 auditor-pairs are associated with lower levels of income-increasing abnormal accruals. Francis et al. (2009, 37) find that in France “firms with one or two Big 4 auditors are less likely to have income-increasing abnormal accruals than other firms. […] firms audited by two Big 4 auditors are

even less likely to have income-increasing accruals.” Big 4 auditors paired with non-Big 4 auditors are also associated with lower levels of income increasing abnormal accruals

however to a lesser extent. Francis et al. (2009) conclude that a pecking order exists with regards to earnings quality and auditor-pair choice. Although we examine the relation between auditor pairs and abnormal accruals, we suggest examining this pecking order from a different perspective, looking specifically at accounting conservatism, in particular timely loss recognition, and considering strategic interactions between auditors.

Zerni et al. (2012) study the impact of voluntary joint audit in Sweden on audit quality. While controlling for differences in characteristics between firms choosing joint audits from other firms, the authors demonstrate that joint audits improve audit quality. Zerni et al. (2012) define audit quality as earnings conservatism, abnormal accruals, credit ratings, and perceived risk of bankruptcy and use a sample of approximately 900 firm-year observations. Our study differs from Zerni et al. (2012) in terms of research objective and context. We focus on the consequences on the auditor pair choice on audit quality in a mandatory joint audit context.

Lesage et al. (2011) tests the impact of joint audit on both audit costs and audit quality in Denmark (2005-2009), which gave up the mandatory joint audit in 2005. Firms that continue to use joint audit after the 2005 regulation change are associated with significantly higher audit fees compared with firms voluntarily choosing to use a single auditor. There is no significant relationship between voluntary joint audit and total fees. In addition, audit quality, proxied by abnormal accruals, is not significantly different for the joint and single audit firms.

Marmousez (2009) examines the consequences of joint auditor pairs in France on financial reporting quality, measured by the degree of earnings conservatism, on a sample of 177 firm-year observations. The author provides evidence that Big 4–Big 4 auditor pairs are not associated with earnings conservatism whereas Big 4–non-Big 4 auditor pairs are associated with conservatism. According to Marmousez (2009), the rationale for these results is that interactions between Big 4 auditors are less efficient and reduce incentives to provide an adequate effort for Big 4 pairs. Our study adds to this exploratory work at two levels. First, relying on a game theory model, we provide a formal background for this rationale. Second, using an alternative proxy for financial reporting conservatism and a larger sample, we test differently the effect on auditor pair choice on timely loss recognition and show consistent results.

We decided to proxy for timely loss recognition using accounting procedures that are essential and likely to be manipulated by managers, i.e. impairment tests. Impairments of assets are typically perceived as a negative asset pricing signal by market participants (Fields

et al. 2001), which provides a primary incentive for firms to avoid booking or delaying impairment of assets. Multiple other incentives exist for managers to avoid or delay impairment recognition including debt and compensation contracts (Watts and Zimmerman 1986) or management reputation (Francis et al. 1996). Consequently Ramanna and Watts (2009) or Li and Sloan (2011) find that impairment of assets are subject to a high degree of manipulation by managers which benefit from a context of subjectivity permissible by the publication of recent accounting standards, i.e. IAS 36 (IASB 2004) internationally and FAS 142 (FASB 2001) in the U.S. Petersen and Plenborg (2010), using a survey on 58 firms listed in the Copenhagen Stock Exchange, identify numerous areas of non-compliance with IAS 36 “Impairment of assets” for these firms and show that “practice varies considerably among firms” (p. 421). The authors also stress that “IAS 36 is a standard that involves substantial judgment” (p. 420). In this context, external auditors play an important role in maintaining the objectivity and fairness of impairment tests, particularly with regard to the accuracy and transparency of impairment-tests (Petersen and Plenborg 2010, 419). The joint audit process may be viewed as the outcome of a non-cooperative game between the two auditors.

Game theory has already been used to describe relations between a firm and its (single) auditor. Strategic relations are analyzed in a cooperative game set (Demski and Swieringa 1974; Hatherly et al. 1996). Hatherly et al. (1996) advance that “the fee structure is (cooperatively) agreed upon” between the auditor and the auditee and that they both share the potential legal costs associated with unacceptable accounting methods. Hatherly et al. (1996) and Antle and Nalebuff (1991) also consider that auditor and auditee jointly agree on a strategy and the auditing process. Other papers suppose non-cooperative interactions between auditor and auditee (Fellingham and Newman 1985; Matsumura and Tucker 1992; Cook et al. 1997). In this setting the auditor can take actions to control the auditee and the conditions for the socially desirable outcome to occur are examined. To our knowledge, game theory has never been used to model strategic interactions between joint auditors.

In the next section we describe our game theory model and derive our hypotheses.

IV. SIMPLE JOINT AUDIT GAME AND DEVELOPMENT OF HYPOTHESES

As managers may be reluctant to book economic impairment, one of the socially desirable outcomes of legal audit is to force managers to conduct objective impairment tests and recognize impairment losses when they occur. The competitive nature of the audit industry

suggest to model joint audit as a non-cooperative game between joint auditors. While investigating impairment tests conducted by managers, joint auditors must decide whether or not to take corrective actions, conditioning the objectivity and transparency of impairment tests. In our model we examine the effect of joint audit on independence as we assume that auditors know that some assets are impaired (competence is supposed to be constant).

Two-Player Game with One Type of Auditor

We hypothesize that joint auditors have the choice between two strategies: (1) taking “corrective actions” to force managers to book economic impairments and be transparent; (2) “no action.” We further hypothesize that:

- Taking corrective actions to force firms to book impairments generates costs CCA for

auditors. These costs are both explicit and implicit: explicit costs consist in the additional hours to be spent to complete the audit process; implicit costs represent the potential harm to the business relation with the auditee. These could be measured as the product of the probability of losing the next tender offer on the legal audit mandate and the fees to earn on this engagement. Harming the relation with the auditee could in certain cases increase the probability of losing mandates or advisory missions. All these costs are supposed to be shared equally between the two auditors (1/2*CCA)

when they both decide to take corrective actions. Auditors bear the full amount of costs CCA when they decide to take corrective actions alone. The costs of “no action”

are set to zero.

- If auditors do not take corrective action impairment tests are not objective and transparent, but one auditor choosing to take “corrective actions” alone is sufficient to force managers to book impairment objectively and be transparent (solutions of the game marked with a star in Table 1).

- Finally, if auditors are associated with non-transparent financial disclosures and managers avoiding economic impairments, they bear reputation costs and potential litigation costs (CREP +CLIT). These costs represent the loss of perceived quality by

clients when observing international audit firms associated with low quality financial statements and the risks associated with the probability of being sued by investors. Those costs are shared equally (1/2*[CREP +CLIT]) between the joint-auditors.

With the same type of auditor, i.e., two Big 4, the players, strategies, and payoffs can be represented in a reduced strategic form as in Table 2.

[Insert Table 2 About Here]

The Nash equilibrium solution for this two-player game will depend on the relative value of the “corrective actions” strategy’s costs as compared to the reputation costs generated by the “no-action” strategy.

Lemma 1: If 1/2*[CREP+CLIT] < CCA then the Nash Equilibrium solution is given when both

firms do not take corrective action similar to the prisoner’s dilemma solution.

Proof: Auditor #1 “no action” strategy dominates when auditor #2 chooses to take “corrective actions” since -1/2 *CCA < 0 and also dominates when auditor #2 chooses to make “no-action”

since 1/2*[CREP+CLIT] < CCA by hypothesis. The same argument holds true for the other

player.

The implication of this result is that if reputation costs and litigation costs are low enough, either because they are low in absolute terms or because they are shared by two Big auditors (both having deep pockets), then the auditors will choose to take “no-action” and impairment tests will be manipulated by managers.8 This is the typical case in the prisoner’s dilemma game, according to which each party is individually better off if the other makes the effort but the socially optimal solution occurs when both make the effort. This is likely to be the case as two Big 4 auditors mutualize their reputation and litigation costs. Therefore we make the following hypothesis:

H1: Firms audited by two Big 4 are more likely to manipulate their impairment tests and have lower levels of conditional conservatism.

In the next paragraph, we introduce two types of auditors. Introducing Big 4 / non-Big 4 Auditor Types

We now consider two types of players: Big 4 and non-Big 4 auditors. We assume the following differences between the two types of auditors:

8

One could also argue that two Big 4 auditors reduce the absolute level of reputation costs since they both signal the quality of financial statements. If a problem occurs later they could argue that it was impossible to detect since they both did not see it.

- Because of economies of scale and human capital, we assume the explicit costs of taking corrective actions to force the auditee to book economic impairment timely is greater for non-Big 4 than for Big 4. We also assume that the implicit costs are greater for non-Big 4 auditors relying on a limited number of mandates to generate revenues. Therefore, for non-Big 4: _ = ∗ _ (with k > 1).9 We hypothesize a linear relationship between these costs for non-Big 4 and Big 4 auditors.10 If both auditors decide to take corrective actions, the costs is reduced for the Big 4 auditor to 1/2*CCA,

and to k/2*CCA for the non-Big 4 auditor.

- The reputation and litigation costs CREP and CLIT generated when an auditor is

associated with manipulated impairment tests are higher for a Big 4 than for non-Big 4 because they have “more to lose” (DeAngelo 1981, 183). Non-Big 4 auditors do not have the financial capacity to face high litigation risks as compared to Big 4 auditors. For the sake of simplicity, we hypothesize that all the reputation and litigation costs fall on the Big 4 auditor and none fall on the non-Big 4 auditor.

With a heterogeneous auditor pair the players, strategies and payoffs can be represented in a reduced strategic form as in Table 3.

[Insert Table 3 About Here]

As in the previous case, the Nash equilibrium solution for this two-person game with two types of auditors will depend on the relative value of the “corrective actions” strategy’s costs as compared to the reputation and litigation costs associated with manipulated impairment tests.

Lemma 2: If CREP + CLIT > CCA then the Nash equilibrium solution is given when Big 4

auditor takes “corrective actions” and non-Big 4 auditor chooses to free ride by taking “no-action.”

Proof: The non-Big 4 auditor strategy of taking “no action” strictly dominates the other strategy since -k*CCA < -k/2*CCA < 0. Given the non-Big 4 strategy, the Big 4 strategy of

taking “correcting actions” dominates the “no-action” strategy if and only if CREP + CLIT >

CCA.

9

We believe that impairment-testing disclosures offer a great setting since they rely on fair value that require specific skills (e.g. intangible assets valuation) that non-Big 4 auditors are less likely to own.

10

The implication of this result is the following: as non-Big 4 auditors bear to a much lesser extent reputation and litigation costs (here none) and face higher costs to implement the corrective actions strategy, they have no incentive to force firms to take impairment, i.e. the strategy “no action” strictly dominates the other strategy. As a result the non-Big 4 auditor acts as a free rider. Given the choice of the non-Big 4 auditor, the Big 4 auditor will choose to take corrective actions only if the reputation and litigation costs associated with manipulated impairment tests (that they fully bear) exceed the costs of taking corrective actions alone.

The costs of taking corrective actions are higher for non-Big 4 firms. These firms are relatively small, the costs of taking corrective actions could be relatively high. For small audit firms, resources are scarce, particularly for those with the kind of expertise required to conduct impairment testing (e.g., valuation skills) explicit costs are higher. In non-Big 4 each legal audit engagement represents a significant share of total revenues of the audit firm. Therefore, the risk of losing a legal audit engagement is usually not acceptable, because it could jeopardize the firm’s financial health. In this context, the implicit costs of efforts are also relatively high for small audit firms. Reputation and litigation costs are higher for a Big 4 auditor in the case of a joint audit with a non-big 4 because Big 4 audit firm have an international reputation to defend and would certainly face litigation by investors in cases of failure. As a result, we make the following assumption:

H2: Firms audited by a Big 4 paired with a non-Big 4 are more likely to be associated with timely and transparent impairments, and more generally more conditional conservatism.

In the next section we test H1 and H2.

V. EMPIRICAL APPLICATION

Impairment Tests: Background

Standard IAS 36 “Impairment of assets” (IASB 2004) prescribes the procedures and disclosures required to perform impairment tests. Standard IAS 36 covers a large range of assets from tangible to intangible assets: lands, buildings, machineries, investment properties, investments in subsidiaries carried at cost, technologies, brands, customer relationships, and goodwill. Impairments are required to be reported in profit or loss if the net book value of an asset is higher than the recoverable value, the latter being the highest of fair value less costs to

sell or value-in-use. If it is impossible to determine the recoverable amount for an individual asset, Standard IAS 36 prescribes to determine recoverable amount for groups of assets known as cash generating units. In order to identify a specific cash generating unit, the associated cash flows must be independent from cash flows arising from other cash generating units. Goodwill is typically allocated to one or several cash generating units and tested indirectly within a cash generating unit or group of cash generating units. The fair value estimate of the cash generating unit is typically based either on a discounted cash flow approach or on a relative valuation method except when a cash generating unit is listed, which is extremely rare. If the recoverable value of a cash generating unit to which goodwill as been allocated is below its carrying value, the entity must recognize an impairment loss. The value of goodwill is written-down first before reducing other assets carrying values.

Because of the valuation methods used, impairments of assets are based on management estimates (Petersen and Plenborg 2010, 420). Managers usually acknowledge that they use specific assumptions for impairment testing purposes. For example in Alcatel-Lucent’s 2008 annual report (p. 245) the management acknowledges that “the recoverable values of our goodwill and intangible assets, as determined for the impairment tests performed by the Group, are based on key assumptions which could have a significant impact on the consolidated financial statements. These key assumptions include, among other things, the following elements: discount rate; and projected cash-flows […].” Managers are often not explicit regarding the valuation assumptions they used to estimate recoverable values

The next section exposes our methodology to examine the relation between auditor pairs and economic impairments.

Auditor Pair and Recognition of Economic Impairment

Impairment tests, in particular impairment tests for goodwill involving discounted cash flow models, require managers assumptions. We examine how auditor pairs affect the decision to recognize economic impairment of assets on the income statement. We argue that according to our game theory model, Big 4–Big 4 auditor pairs will be more likely to avoid booking economic impairment in a timely fashion, whereas Big 4–non-Big 4 auditors pairs will be related with more timely impairments. Because they share reputation and litigations costs, Big 4 paired with other Big 4 could lead more often to the avoidance of impairments, when impairments should be booked. On the opposite, Big 4 paired with non-Big 4 should be

related with the timely recognition of impairments because the Big 4 auditor bears all reputation and litigations costs and will force firms to recognize impairment losses.

As the existence of an economic impairment of asset is not directly observable, we consider that economic impairments are likely if firm’s return on assets (EBITDA divided by total assets) is below a given threshold, i.e. below the 1st quartile return of assets of our sample.11 In order to test our hypotheses, we estimate the following probit model:

Pr (,) = ( +  !"#$%&_(#&,+ )*%+,&,

+ -*%+,& ∗ !"#$%&_(#&,+ .#/,,+ 012,+ 3$% , + 4,&,+ 56,$(,+ 78200+ ;)

(1)

Where for firm i and year t:

- DIMPi,t = 1 if the firm books an impairment of assets (Annual report);

- Auditor_Pairi,t = one of the two following dummy variables:

- Big4_Big4i,t = 1 if both firm’s external auditors are Big 4 auditors (Annual

report);

- Big4_Smalli,t = 1 if one of the two external auditors is a Big 4 auditor and the

other is not (Annual report);

- LowPerf = 1 if the firm return on asset (EBITDA divided by total assets) is lower than the 25th percentile of the sample, i.e. below 7.3% (see Table 4, Panel B) (Datastream); - Sizei,t = natural logarithm of total assets (Datastream);

- GWi,t = goodwill divided by total assets (Datastream);

- MtoBi,t = market-to-book ratio of equity (Datastream);

- Perfi,t = return on assets (EBITDA divided by total assets) (Datastream);

- Ubetai,t = 5-year unlevered beta (Datastream);

- Y200X = 1 for year 200X.

If H1 is true, we expect a negative relation between the probability to book an impairment and a firm having two Big 4 auditors when performance is low, i.e. a statistically significant negative sign for the coefficient of variable LowPerf*Big4_Big4. According to hypothesis 1, firms facing an economic impairment and having a Big 4 paired with another Big 4 tend to avoid the recognition of economic impairments. On the opposite, we expect a positive

11

We also estimated our model with other threshold for low performance, i.e. 5th percentile, 10th percentile, 15th percentile and 20th percentile. Results are qualitatively similar.

association between a firm having a Big 4 paired with a non-Big 4 auditor and the probability to book an impairment when performance is low, i.e. a statistically positive sign for coefficient on variable LowPerf*Big4_Small. According to hypothesis 2, firms audited by a Big 4 paired with a non-Big 4 take economic impairments on the income statement.

We control for performance, size, goodwill balance, market-to-book ratio, risk and years. We expect that the likelihood of impairments decreases with Performance (Perf), market-to-book ratio (MtoB), and size (Size). We anticipate that the likelihood to recognize an impairment increases with goodwill balance (GW), risk (Ubeta) and during the financial crisis of 2008.

In the next paragraph, we examine the relation between auditor pair and transparency of impairment tests.

Auditor Pair and Transparency of Impairment Tests

Measuring Transparency of Impairment Tests

For French listed firms, transparency regarding impairment tests varies greatly from one firm to another. For instance the French pharmaceutical company Stallergenes in its 2006 annual report’s “Main Accounting Methods” section (p. 41), only provides minimal narrative information with regard to impairment tests such as “A writedown is recorded once a year or more frequently if events or changes in circumstances indicate the likelihood of impairment for that acquisition goodwill” and “If an impairment is identified, the recoverable value of the CGU to which the acquisition goodwill belongs is assessed. An impairment is recognized as soon as the book value of the CGU to which the acquisition goodwill belongs exceeds the recoverable value.” No further information concerning impairment tests is provided in the Notes, although the firm owns a substantial amount of intangibles assets for which impairment tests are required to be performed at least once a year: goodwill alone represents 24% of Stallergenes’ total assets.

On the opposite, France Telecom’s 2008 annual report contains are much more transparent regarding impairment-testing procedures. In Note 6 (p. 287-289), the company dedicates almost three pages to its impairment tests and provides a wide range of information. The management explains the level at which goodwill is tested, and provides tables with key assumptions used in the estimation of recoverable amounts (e.g., growth rate to perpetuity,

main cash generating units and groups of cash generating units, post and pre-tax discount rates used), as well as narrative explanations for specific countries where it operates.12

Considering these differences in transparency and the incentives to manipulate impairment tests, we examine every annual report produced by French non-financial 120 top-listed firms (SBF 120 index) from 2006 to 2009. We look for 40 items covering the main disclosures required by IAS 36 (paragraphs 126 to 133) and other disclosures providing the main valuation assumptions used in the estimation of recoverable values. Our 40 items are allocated to categories of information according to homogeneous topics. Appendix 1 exhibits the main categories, and items in each category covered by our impairment test Transparency Score.

[Insert Table 4 About Here]

By attributing one point per item that shows up in firm i’s annual report for year t, the resulting Transparency Score is computed as:

.%&,, =_{40 = $,>}1 ,∗ 100 ?

7@

(2)

We divide the sum of the items by the maximum number of points and multiply it by 100 in order to obtain the dependent variable Score ranging from 0 to 100 capturing transparency of impairment tests.

As presented in Appendix 1, our Transparency Score is quite comprehensive covering technical valuation elements of impairment tests (e.g., discount rates, neutrality of the financing structure, terminal value issues) as well as descriptive aspects (e.g., presentation of the alternative fair value or value-in-use, use of independent experts).

In the next paragraph we explain our methodology to test the relation between transparency of impairment tests and auditor-pair choice.

Auditor Pair and Transparency of Impairment Tests: Model

To determine if the auditor pair has an impact on the level of transparency, we estimate model (3), according to which Auditor_Pair and DIMP*Auditor_Pair are the main variables of interest:

12

These two examples have been selected on purpose from a low disclosing firm (Stallegenes, 2006) and a high disclosing firm (France Telecom, 2008), based on our Transparency Score.

.%&,, =  +  !"#$%&_(#&,+ ),+ -,∗ !"#$%&_(#&, + AB%($,

+ 12,+ 06,$(,+ 3.#/,,+ 4*,C,&(D,,+ 5ℎ(FD,,

+ G%C,&(D,,+ =  ?H78200+ ;, (3)

Where for firm i and year t:

- Scorei,t = self-constructed Transparency Score for firms’ impairment tests (Annual

report);

- Auditor_Pairi,t= one of the following two dummies variables:

- Big4_Big4i,t = 1 if both firm’s external auditors are Big 4 auditors (Annual report);

- Big4_Smalli,t = 1 if one of the two external auditors is a Big 4 auditor and the

other is not (Annual report);

- DIMPi,t = 1 when the firm recognizes an impairment of asset (Annual report);

- Floati,t= percentage of firm’s publicly exchanged equity (free float) (Datastream);

- GWi,t = goodwill divided by total assets (Datastream);

- Ubetai,t = firm’s 5-year unlevered beta (Datastream);

- Sizei,t = natural logarithm of the firm’s total assets (Datastream);

- Leveragei,t = Leverage is the firm financial debt minus cash and cash equivalents,

divided by the market value of equity (Datastream);

- Changei,t = 1 if the firm changed external auditor during the fiscal year (Annual

report);

- Coveragei,t= natural logarithm of the number of recommendations issued by financial

analysts during the year (I/B/E/S); - Y200Xt = 1 for year 200X.

We anticipate that the unconditional level of transparency is higher for Big 4–Big 4 pairs than for other combinations, but that Big 4–Big 4 pairs manage the level of transparency. According to H1, we anticipate that when they book (presumably untimely) impairments, Big 4–Big 4 auditor pairs reduce their level of impairment test transparency as compared to Big 4–non-Big 4 auditor pairs. We anticipate a negative relationship between Score and

DIMP*Big4_Big4, because Big 4 auditors paired with other Big 4 auditor manage the level of impairment-testing transparency as they book untimely impairments. Conversely positive relationship is expected between Big4_Small and Score when firms recognize an impairment as Big 4–non-Big 4 auditor pairs increase the level of transparency when firms book (timely) impairments (consistent with our model predicting H2).

A positive relationship is expected between Transparency and Float as outside investors are the primary users of annual reports from which our proxy for transparency is derived. This increases the likelihood of finding additional disclosures in the annual report when free float is large. We anticipate a positive association between Score and DIMP as the existence of impairments could trigger additional disclosures explaining the impact and potential reasons for this bad news. We expect a positive association between goodwill balance (GW) and transparency of impairment tests given the importance of impairment test for goodwill. A positive relation is expected with Ubeta as risky firms could attempt to reduce the external perception of their risk by producing additional disclosures. We predict a positive association between Score and Size as firm’s size allows to dedicate more resources to the production of financial disclosures. A positive relation between Leverage and Score is anticipated as highly leverage firms would face additional pressures to disclose more for debt-holders. We expect a negative relation between Change and Score because one or two new auditors would need time to achieve the same level of quality than the previous auditor(s) due to learning effects.

Coverage captures an alternative channel of communication. Everything else equals, a firm followed by more analysts could disclose less information in its annual report as analysts can substitute the annual report (Botosan 1997, 326). We expect a negative relationship with

Score.

Auditor Pair and Conservatism: Standard Models Testing Unconditional and Conditional Conservatism

We also test the effect of auditor pairs on conservatism using standard measures of conservatism from the empirical literature. We differentiate between unconditional and conditional conservatism.

Unconditional conservatism (also known as ex ante or news-independent conservatism) consists in (continually) understating the book value of net assets relatively to their economic value. This form of conservatism, which is an accounting bias toward reporting low earnings and book values of stockholders equity, leads to higher (internally generated) goodwill and higher market-to-book ratio. Unconditional conservatism is a primary (though not the sole) source of unrecorded goodwill, which also captures the present value of expected economic profits (rents, growth). Empirical proxies for unconditional conservatism used in the literature are theoretically based on the Ohlson (1995) residual income model. Roychowdhury and Watts (2007) and García Lara and Mora (2004) use the market-to-book ratio to proxy for

unconditional conservatism. Roychowdhury and Watts (2007) argue that the market-to-book ratio is influenced by two factors: (1) the unverifiable (unbooked) increases in value of separable assets in place (true unconditional conservatism), and (2) the expected value of economic profits (e.g., synergies between asset in place, growth, rents). Therefore it is necessary to control for factor (2) using variables such as the intangibility of the business, firm growth potential, current performance, risk and volatility and investment activity, we estimate the following model adapted from Piot et al. (2011):

$% , = I?+ I #D4_ #D4,+ I) #D4_.>(BB,+ I-12,+ I6,$(,

+ I0J,+ I3∆.(B,L#,$+ I₇N,$!&F#,$+ I₈,&_#,$+= 78200+ P

;, (4)

Where for firm i and year t:

- Big4_Big4i,t = 1 if both firm’s external auditors are Big 4 auditors (Annual report);

- Big4_Smalli,t = 1 if one of the two external auditors is a Big 4 auditor and the

other is not (Annual report);

- GWi,t = goodwill divided by total assets (Datastream);

- Ubetai,t = firm’s 5-year unlevered beta (Datastream);

- PPEi,t = property, plant, and equipment divided by total assets (Datastream);

- ∆Salei,t= current year sales growth (Datastream);

- Returni,t= current year annual share return (Datastream);

- Perfi,t = return on assets (EBITDA divided by total assets) (Datastream);

- Y200X = 1 for year 200X.

We expect that firms audited by a Big 4–non-Big 4 pair are associated with higher market-to-book ratio because book value is more understated than book value of firms audited by a Big 4–Big 4 auditor pair.

Conditional conservatism (also known as ex post or news-dependent conservatism) consists in writing down book values and decreasing income under sufficiently adverse circumstances whereas book value is not written up when circumstances are favorable. We use the Basu (1997)’s piece-wise linear asymmetric model:

, = I?+ I ,+ I)N, + I- ,N,+ ;, (5) Where for firm i and year t:

- Xi,t = earnings per share scaled by share price at the beginning of the fiscal year

- Ri,t = Share return measured from nine months prior to fiscal year-end to three

months after fiscal year-end (Datastream); - BNi,t = 1 if share return is negative.

We expect that

β

4–non-Big 4 pair than for firms audited by a Big 4–Big 4 pair.

Auditor Pair and Quality of Earnings

We use discretionary accruals to proxy for accrual-based earnings management. Discretionary accruals are the difference between firms’ actual accruals and the normal level of accruals. We estimate the following modified Jones (1991) model:

&!(BL,/R,S = T?+ T (1/R,S ) + T)(∆./R,S ) + T-(J,/R,S ) + ,, (6) Where for firm i and year t:

- Accrualsi,t = earnings before extraordinary items and discontinued operations minus

operatings cash flows (Datastream);

- ∆. = change in sales from year t-1 to t (Datastream); - R_,S = total assets of the previous year (Datastream); - PPEi,t = gross property, plant, and equipment (Datastream).

We estimate model (6) per industry with at least 15 firm-year observations. The residuals of model (6) are abnormal accruals (AbAccruals).

To assess the relation between earnings quality and auditor pairs we estimate the following model:

&!(BL, = T?+ T #D4_ #D4,+ T) #D4_.>(BB,+ T-,&,+ T*,C,&(D,,

+ T0.#/,,+ ;

(7)

Where for firm i and year t:

- AbAccrualsi,t= residuals from model (6);

- Big4_Big4i,t = 1 if both firm’s external auditors are Big 4 auditors (Annual report);

- Big4_Smalli,t = 1 if one of the two external auditors is a Big 4 auditor and the

other is not (Annual report);

- Perfi,t = return on assets (EBITDA divided by total assets) (Datastream).

- Leveragei,t = Leverage is the firm financial debt minus cash and cash equivalents,

- Sizei,t = natural logarithm of the firm’s total assets (Datastream).

Model (7) tests the impact of auditor pairs on earnings quality and controls for performance, size and financial leverage. Literature shows that Big 4 auditors improve earnings quality, in particular reduce income-increasing earnings management (Francis et al. 2009).

In the next section we present our sample and results.

Sample and Empirical Results

We start our sample with the 120 largest listed firms composing the SBF 120 index over the 2006 – 2009 period. We remove the 10 financial firms from our sample due to the specificities of the industry’s impairment and disclosures. Due to missing variables for some firms, our final sample is composed of 91 firms representing 325 firm-year observations and 10 industries (see Table 4, Panel A). All continuous variables are winsorized at 1%.

[Insert Table 4 About Here]

From Table 4, Panel B, it appears that approximately 44% of the firms are audited by a Big 4–Big 4 auditor pair over the period, 54% of firms are audited by a Big 4–non-Big 4 auditor pair and the remaining 2% are audited by a pair of non-Big 4. These statistics illustrate that the French audit market is one of the least concentrated as more than one firm out of two are audited by at least one non-Big 4 auditor.

Impairment occurred on average 44% of the time during the period. Table 4, Panel C, shows that firms audited by Big 4–Big 4 auditor pairs booked impairment more often than other auditor pairs unconditionally to performance: 52% of the time when the mean is only 44% (difference significant at less than 1% one-tailed). In comparison, firms audited by Big4–non-Big 4 auditor pairs booked less often impairments unconditionally to performance, i.e. 35% of the time (difference significant at less than 1% one-tailed). However, Table 4, Panel D, presents evidence that conditionally to the existence of an economic impairment, when market-to-book ratio is below one, firms audited by Big 4–Big 4 auditor pairs booked

economic impairment only 45% of the time whereas Big 4–non-Big 4 auditor pairs booked impairment 67% of the time (difference significant at less than 6% one-tailed). This first result is consistent with H1 and H2 in the sense that firms audited by Big 4–non-Big 4 auditor pairs show higher timely loss recognition: they book impairment when they occur.

From Table 4, Panel B, the mean (median) impairment-test Transparency Score for the four-year period is 54 points (54 points) and range from a minimum of 13 points to a maximum of 89 points over the period. From Table 4, Panel E, we see that transparency increases through time, with a mean (median) score rising from 50 pts (51 pts) in 2006 to 59 pts (58 pts) in 2009. From Table 4, Panel F, the unconditional level of firms’ transparency is higher for Big 4–Big 4 pairs than for other auditor pairs (significant at less than 1% one-tailed). The average transparency is higher for pairs of Big 4 auditors in comparison with other pairs. However, from Table 4, Panel G, the transparency of impairment tests conditional to the occurrence of impairments is not significantly different for Big 4–Big 4 than for other auditor pairs as Big 4–non-Big 4 pairs exhibit higher transparency and Big 4–Big 4 slightly lower transparency (t-tests fail to reject the null of no difference). This result suggests that on average firms audited by Big 4–Big 4 auditor pairs are more transparent but their transparency does not increase or tend to decrease when they book (presumably untimely) impairments. On the contrary, Big 4–non-Big 4 auditor pairs increase their transparency when they recognize (presumably timely) impairments. These results are consistent with our hypothesis that Big 4– Big 4 manage the transparency of their tests to avoid booking economic impairment.

The dispersion of the firms’ scores in the sample is relatively high and firms improve their transparency over the period 2006-2009. The dispersion and evolution of the Transparency Score is exhibited in Figure 1.

[Insert Figure 1 About Here]

Table 4, Panel H, shows that market-to-book ratio is significantly higher for firms audited by a Big 4–Small pair (significant at less than 5% one-sided), consistent with Big 4–non-Big 4 leading to more unconditionally conservative financial statements. Conversely, Big 4–Big 4 pairs are associated with lower market-to-book ratio (significant at less than 5% one-sided), potentially indicating overstated assets.

From Table 4, Panel B, we observe that changes of either one or two auditors during a year do not occur frequently, i.e. only 4% of the time. The dispersion of the score as measured by the standard deviation is more than 16 points. The mean (median) risk of firms in the sample as proxied by 5-year unlevered beta is 0.99 (0.74), the mean (median) size is 15.7 (15.6). The mean (median) percentage of firms’ shareholder’s equity that is available to trade (free float) is 67% (67%). Impairment of assets represent on average 2% of EBITDA over the time

period. The average (median) performance measured by return on assets (EBITDA divided by total assets) is 12.4% (10.8%) and the average market-to-book ratio is 2.29 (1.98).

Table 4, Panel G, shows the correlation between variables. DIMP is significantly positively correlated with Big4_Big4, Size, Small_Small, and Coverage; and negatively correlated with

Big4_Small and MtoB. We can see that Score is positively correlated with Size, Float, GW,

Leverage and Coverage which is consistent with our expectations expect for Coverage. Correlation between Score and Big4_Big4 is positive and statistically different from zero (significant at 1% two-tailed) confirming that Big4–Big 4 auditor pairs are on average unconditionally more transparent about impairment tests. The level of transparency is also negatively correlated with the absolute value of abnormal accruals (significant at less than 2% two-tailed) suggesting that lower earnings quality is related with lower transparency.

We present the results of our multivariate analysis in the next paragraph. Auditor Pair and Recognition of Economic Impairments

Model (1) examines the association between auditor pairs and timeliness of impairments. Table 5 reports the results of the OLS estimation with robust t-statistics.

[Insert Table 5 About Here]

Table 5 shows the marginal effects of model (1). Big4–Big 4 auditor pairs appear to impair assets more often than Big4–non-Big 4 auditor pairs, independently from firms’ performance (see marginal effect b1 = 0.18, significant at less than 2% two-tailed). This case is consistent

with impairment-testing management through smoothing impairment loss through time. Pairs of Big 4 auditors allow audited firms recognizing impairments a little at a time. However, when performance is low, firms audited by a pair of Big 4 auditors are also more likely to avoid taking economic impairments (see marginal effect of LowPerf*Big4_Big4 = -0.31, significant at less than 1% two-tailed). This is also consistent with impairment-testing management because economic impairment losses are not booked when they occur, i.e. when return on assets is low. On the contrary, firms audited by Big 4 paired with non-Big 4 auditors recognize impairment losses less often, unconditionally to performance (marginal effect of variable Big4_Small = -0.23, significant at less than 1% two-tailed). However, when performance is low, they are more likely to recognize economic impairment losses. The marginal effect for LowPerf*Big4_Small = 0.32 is positive (significant at less than 1% two-tailed). This is consistent with timely loss recognition and an effective control of auditors on

impairment tests. These results confirm our predictions derived from our game theory model and what we observe in the descriptive statistics of Table 4, Panel D.

The next paragraph examines the relation between transparency and auditor pairs. Auditor Pair and Transparency of Impairment Tests

The association between auditor pairs and transparency of impairment tests is tested with model (3). Table 6 reports the results of the OLS estimation with robust t-statistics.

[Insert Table 6 About Here]

Table 6 indicates that when firms take an impairment, they tend to reduce the level of transparency when they are audited by a Big 4–Big 4 auditor pairs in comparison with other pairs of auditor (Coefficient on DIMP*Big4_Big4 negative and significant at less than 7% two-tailed). We observe the opposite results when examining the relation between transparency and Big 4–non-Big 4 auditor pairs: firms become more transparent when they book impairments (Coefficient on DIMP*Big4_Small positive and significant at less than 10% two-tailed).

Nevertheless, these results are consistent with our game theory model predictions, i.e. Big 4 auditor pairs sharing the reputation costs are more likely to engage in manipulation of impairment-testing transparency to support untimely impairment. Conversely, Big 4 auditors bearing all reputation and litigation costs when paired with non-Big 4 auditors face strong incentives to force firms to higher levels of transparency when they recognize impairments. However, considering the low significance of these results, they must be considered only as weak evidences consistent with our predictions.

From the control variables presented in Table 6, it appears that goodwill balance, and

Size significantly increase the level of transparency (confirming the expected relationships).

Change is negatively associated with the level of transparency, consistent with the expected relationship. The transparency is increasing through time as the coefficient on the year dummies rises for years 2008 and 2009 confirming what was observed on Figure 1.

Coverage, Leverage and Float variables do not appear to be significantly related with impairment-testing transparency. The model explains approximately 20% of the variance of the Transparency Score.

Auditor Pair and Conservatism: Standard Measures

Table 7, Panel A, shows the estimation results of model (4) that examines the association between auditor pair and unconditional conservatism. Table 7, Panel A, reports the results of the OLS estimation with robust t-statistics.

[Insert Table 7 About Here]

Controlling for future economic profits with intangibility of the balance sheet (percentage of total assets composed of goodwill and of PPE), current performance (share return and sales growth) and risk (beta coefficient), Big 4 auditors paired with Small auditors appear to be significantly associated with higher levels of market-to-book ratios (significant at less than 1% two-sided). This indicates stronger level of unconditional conservatism, i.e. lower levels of book values. We do not find evidence of significantly higher level of market-to-book ratio for firms audited by a Big 4–Big 4 pair. This result is consistent with Big 4–non-Big 4 being more conservative independently from any economic losses.

Table 7, Panel B, presents the estimation results of the Basu (1997)’s model estimated with OLS and reports robust t-statistics. The model is estimated on the total sample and on two sub-samples: (1) firms audited by a Big 4–Big 4 pair, and (2) firms audited by a Big 4–non-Big 4 pair. The degree of conditional conservatism, i.e. the asymmetric treatment of bad news vs. good news, is captured by

β

significant coefficient

β

gains.This is only the case in the sub-sample of firms audited by a Big 4–Small auditor pair: the coefficient is significant (at less than 1% two-sided) and is larger than the coefficient estimated on the sub-sample of firms audited by a Big 4–Big 4 pair (the coefficient

β

smaller and not significant). This is consistent with our predictions and confirms the results observed using impairment tests. Interestingly, the coefficient

β

good news into earnings, appears small and significantly negative for the total sample and the two sub-samples. We believe that this unexpected result may be due to the time period of the sample, i.e., 2006-2009 that has seen a major financial crisis.

Auditor Pair and Earnings Quality

Model (7) examines the relations between auditor pairs and the quality of earnings – measured by the level of abnormal accruals in earnings. Table 8 presents the results of models (6) and (7).